A study on DCT coefficients of time series pen position data for printed signature verification

For the purpose of secure signature verification, it is preferable to use small sized feature data which is stable between genuine signatures and difficult to be mimicked by illegal transactions. Although the turning points extracted from XY position and pressure of pen is one of efficient feature data, they are not suitable for printed signatures written by Chinese or Japanese. This is because the signature consists of a number of segmented straight lines and has a fewer turning points than those of Western signatures. In order to reduce data size of printed signature, we have applied DCT (Discrete Cosine Transform) coefficients of time series XY position to signature verification. However, the DCT spectrum distance between genuine signatures acquired in different times is not small enough to accept. In this paper, we point out that the DCT spectrum distance of genuine signatures after DP matching becomes tremendously small and DCT spectra in low frequency domain has essential individuality. From experimental results using 20 Japanese persons, the 30 DCT spectra of Y position verifies the individuals with 1.7% EER(Equal Error Rate). This EER is similar with the results obtained by the signature verification system in which whole series data of XY position and pressure is used.